1Source-Language Features and Maximum Correlation Training for Machine Translation Evaluationconstrained n-gram precision. Overlapping n-grams between an MT hypothesis and its references do not necessarily indicate correct...
2Extending the METEOR Machine Translation Evaluation Metric to the Phrase Leveltranslation hypotheses. A human annotator edits each MT hypothesis so that it is meaning-equivalent with a reference translation...
3Incremental Adaptation of Speech-to-Speech Translation2009 - Nguyen Bach,Roger Hsiao,Matthias Eck,Paisarn Charoenpornsawat,Vogel Stephan,Tanja Schultz,Ian Lane,Alex Waibel,Alan W. Blackbaseline ASR hypothesis and use the baseline SMT to get the MT hypothesis and a 500-best list. We train a domain LM using the 500-best...